I finally made a waterproof sensor that reads in PAR units, but reads about 10% low. I ended up using this design:
It is made from a piece of 3/4 inch diameter acrylic tube, an acrylic half dome, a strip of 1/8 inch thick acrylic, for a handle. First I cut a piece of tube about .02-.03 longer than the photodiode with its leads - Excelitas VTB8441BH - from Newark Electronics, about 3/8 inches long. Then I drilled a hole in one side at about the height that I expected the electric cable to come in, just big enough to get the cable through it. I had a 1" by 5" piece of acrylic !/8 inch thick sheet, so I glued that tube to that with Weldon #16. When that dried I used black nail polish to paint the inside of the tube a solid black.
To solder the cable to the photodiode I drilled a shallow 5/16 inch diameter hole in a scrap of plywood. Then I jammed the photodiode in that holes with the leads up. This acted as a third hand to hold it when I wrapped the two cable wires around the leads and soldered them on.
To get the diode inside the tube I poked the other end from inside the tube, through the hole in the side, and pulled it until the diode could fit into the tube. I centered the diode, with the ends of the leads against the bottom of the tube enclosure, and used masking tape to hold the cable steady against the bottom piece.
I used Loctite instant mix, 5 minute setting epoxy to form a pool of epoxy about 1/8" deep, deep enough to cover the cable hole in the tube, to anchor the photodiode in place.
Then the hard part began: selecting Roscolux filters to convert the reading to PAR. First I wanted to change the wave length sensitivity from a peak at about 600 nm to something more nearly flat over the 400-700 nm range. At the same time I needed to greatly reduce the sensitivity of the photo diode. This required a spreadsheet calculator, with all of the transmissivity data for lots of filters, by wave length. First I found a filter to knock off the 600 nm peak, without too much damage to the rest of the sensitivity. Then a filter to correct, as much as I could, the damage that was done to the rest of the sensitivity. (This means first reading the graph of sensitivity for the diode at every 20 nm and putting that in the spreadsheet, then multiplying all of the filter sensitiviies by the diode sensitivity, and dividing by the resulting maximum peak sensitivity to get a graph with the peak at 1.0.) To act as a diffuser to emulate the cosine diffuser, I used a Roscolux diffusing filter, and multiplied those sensitivities in with the others. When I had the best looking spectral distribution I could get, which took several tries, ending up with 5 filters, I cut half inch diameter discs of each and stacked them on top of the photodiode, and sat the hemishpere on top, and measured the output with my "standard" light. I got 1112 vs the 25 I wanted. So, I had to find a stack of neutral filters that would drop that 1112 down to 25. That took another 6 filters.
After a final trim of the stack of 11 filter discs I restacked them on top of the photodiode and laid the hemisphere on top and repeated the reading - I got 22 this time, which was close enough, since no other combination was nearly that close.
The filters I used are: #163, #4915, #4230, 2 #R09, #97, 2#397, #98,#398, and #3318. Unfortunately there is no Rosco filter in the sample book that cuts out the infrared and near infrared.
By now the acrylic cement and the epoxy had set up very well, so I glued the hemisphere on top with Weldon #16, and put a filet of that cement around the electric cable on the outside to better waterproof it.
After it set up good, I put a bead of Weldon #16 around the joint at the hemisphere to tube interface to better seal that.
One more calibration, and it still read 22, where 25 would be the desired reading. A final calculation of the sensitivity vs wavelength gave me:
It looks pretty ratty, but it should still work ok. Eventually I will try it at noon in full sun, to see if the 10% error is the same at high light.
Much of the cost of this was in Epoxy and nail polish! About $9 for them, plus $4.26 for the photodiode and about $7 for shipping.